AI agents are extremely pro-active. If they run for long enough or with generic enough prompts they will just do shit for the sake of continue running and doing things following a unachievable goal.